Volume 169, Issue 7 p. 1430-1446
REVIEW
Free Access

Efficacy and ligand bias at the μ-opioid receptor

E Kelly

Corresponding Author

E Kelly

School of Physiology and Pharmacology, University of Bristol, Bristol, UK

Correspondence

E Kelly, School of Physiology and Pharmacology, University of Bristol, Bristol BS8 1TD, UK. E-mail: [email protected]

Search for more papers by this author
First published: 03 May 2013
Citations: 109

Abstract

In order to describe drug action at a GPCR, a full understanding of the pharmacological terms affinity, efficacy and potency is necessary. This is true whether comparing the ability of different agonists to produce a measurable response in a cell or tissue, or determining the relative ability of an agonist to activate a single receptor subtype and produce multiple responses. There is a great deal of interest in the μ-opioid receptor (MOP receptor) and the ligands that act at this GPCR not only because of the clinically important analgesic effects produced by MOP agonists but also because of their liability to induce adverse effects such as respiratory depression and dependence. Our understanding of the mechanisms underlying these effects, as well as the ability to develop new, more effective MOP receptor drugs, depends upon the accurate determination of the efficacy with which these ligands induce coupling of MOP receptors to downstream signalling events. In this review, which is written with the minimum of mathematical content, the basic meaning of terms including efficacy, intrinsic activity and intrinsic efficacy is discussed, along with their relevance to the field of MOP receptor pharmacology, and in particular in relation to biased agonism at this important GPCR.

Linked Articles

Recent reviews on aspects of efficacy can be found at:

Kenakin, T (2013). New concepts in pharmacological efficacy at 7TM receptors: IUPHAR Review 2. British Journal of Pharmacology 168: 554–575. doi: 10.1111/j.1476-5381.2012.02223.x

Roche D, Gil D and Giraldo J (2013). Mechanistic analysis of the function of agonists and allosteric modulators: reconciling two-state and operational models. British Journal of Pharmacology 169: 1189–1202. doi: 10.1111/bph.12231

Abbreviations

  • DAMGO
  • [D-Ala2, NMe-Phe4, Gly-ol5]-enkephalin
  • β-FNA
  • β- funaltrexamine
  • RGS
  • regulator of G protein signalling
  • Introduction

    When discussing data regarding opioid receptor signalling with other biomedical scientists or clinicians, particularly those with no formal pharmacological training, it is noticeable how often they misunderstand the meaning of terms such as efficacy and receptor reserve. These are important concepts in light of the increasing efforts to identify biased ligands for opioid receptors and other GPCRs because the accurate determination of efficacy is essential for the detection of ligand bias. Taking the clinically important analgesic drug morphine as an example, the fact that this drug can readily induce some cell signalling responses, yet is poor at inducing μ-opioid receptor (MOP receptor) internalization, is taken by many as evidence that morphine is fundamentally different from other MOP receptor agonists (receptor nomenclature follows Alexander et al., 2011). However, as will be discussed later, such apparently striking differences may, in some cases, be explained by more mundane factors, such as differing receptor reserves for distinct responses generated by an agonist at a single receptor subtype. Another issue is that some of the research articles and reviews covering efficacy and receptor reserve can be fairly impenetrable for those who have limited pharmacological or mathematical experience. In this review, and beginning with the simpler concepts of receptor theory, I have attempted to explain what efficacy is, how efficacy at MOP receptors or any other GPCR can be measured and how efficacy can then be used to identify biased ligands. The initial sections of this review cover basic material, and the reader who is familiar with these concepts may like to skip these paragraphs.

    The idea of efficacy

    The potency of an agonist in producing a response, usually represented by its EC50 value (Figure 1A), is a function of both the affinity and the efficacy of the agonist at the receptor. Efficacy itself can be described as the ability of a drug, once bound to a receptor, to activate the receptor and produce a cellular response. However, the measurement of drug efficacy is not always straightforward and, for full agonists, efficacy cannot be read directly from a concentration–response curve. This is because, although the maximum response that an agonist produces is related to its efficacy, the maximum response is also often tissue-limited. A tissue limit could be imposed by the concentration of G protein or effector enzyme present. Also, tissue limits occur where, for example, an agonist induces complete inhibition of a response such as transmitter release or muscle contraction before maximum activation of the receptors has occurred. In addition, with opioids, other factors may influence the maximum response measurable. With analgesia testing in rodents, for example, ethical considerations limit the amount of time a painful stimulus can be applied, with the result that many MOP receptor agonists considered to have relatively low efficacy at the receptor, appear as full agonists in these assays (Madia et al., 2009; Hull et al., 2010). Consequently, for the various reasons described previously, the maximum response of an agonist in a tissue can, in some cases, be reached by occupying only a fraction of the available receptors present in the tissue. In the case illustrated in Figure 1B, three of the agonists produce the same tissue maximum response, but they could have quite different efficacy values. In the tissue where the experiment was undertaken, these agonists would be referred to as ‘full agonists’, and the reason why they have different potencies (i.e. different EC50 values) is because they have different affinities, different efficacies or both. In other words, because potency depends upon both affinity and efficacy, the agonist depicted in red in Figure 1B is the most potent of the full agonists for the response because its combination of affinity and efficacy is greater than the other two full agonists.

    figure

    (A) Agonist action is usually shown as a log concentration–response curve, with the curve being defined by the agonist potency (EC50 value), the agonist maximum response (Emax; this may or may not be the tissue maximum) and also by the slope at the midpoint of the curve. (B) Diagrammatic representation of a family of agonist concentration–response curves produced by activation of a single receptor subtype in a tissue. Note that since partial agonists have low efficacy for the response measured, then the partial agonist represented by the grey curve in (B) must have a very high affinity in order to have such a high potency (i.e. low value of EC50). Buprenorphine activation of [35S]GTPγS binding would be an example of such a MOP receptor agonist (McPherson et al., 2010).

    In some cases, agonists have such low efficacy that they cannot achieve the maximum response that a full agonist does, even when occupying all the receptors present in the tissue (Figure 1B); these agonists are called partial agonists. For example, with MOP receptors, ligands such as buprenorphine, meperidine and pentazocine behave as partial agonists in many cell signalling assays (McPherson et al., 2010). The idea that agonists can produce a maximum response in a tissue by occupying only a fraction (less than 100%) of the available receptors leads to the idea of receptor reserve. An agonist with high efficacy will thus have a greater receptor reserve than one with lower efficacy, and an agonist that does not have sufficient efficacy to produce the tissue maximum (i.e. a partial agonist) has no receptor reserve. For example, if one of the full agonists in Figure 1B was able to produce the maximum tissue response by occupying 8% of the available receptors in the tissue, then the receptor reserve would be 92%; if another of the full agonists did so by occupying 95% of the available receptors, then the receptor reserve would only be 5%. Even at the highest concentrations, the partial agonists can only produce a maximum response of less than 100% of the tissue maximum, so have receptor reserves of zero.

    One way to investigate the presence of a receptor reserve is to inactivate a fraction of receptors in a tissue with an irreversible receptor antagonist (Nickerson, 1956) to see how this affects the agonist concentration–response curve. This approach has also been used for MOP receptors, with irreversible MOP receptor inactivators such as β-funaltrexamine (β-FNA), β-chlornaltrexamine and clocinnamox (Chavkin and Goldstein, 1982; Williams and North, 1984; Mjanger and Yaksh, 1991; Chan et al., 1995). If irreversible inactivation of a fraction of receptors is performed for two full agonists that have the same efficacy, then the concentration–response curves will be affected in the same way; if they have different efficacies, then the curves will behave differently following receptor inactivation. In the example shown in Figure 2A, agonists R and B have similar potency values and produce the same maximum response, but R has higher efficacy than B. Pretreatment with a low concentration of an irreversible antagonist shifts the curve for agonist R to the right but does not affect the maximum response to the drug, while for agonist B, there is little shift, but the maximum response decreases markedly (Figure 2B,C). With a higher concentration of irreversible antagonist to inactivate more receptors, the maximum for agonist R now decreases, while the curve for agonist B collapses almost completely. The reason why R initially shifts to the right with no drop in maximum is that there are still sufficient receptors left for agonist occupation to give the maximum response; however, as there are fewer receptors left after inactivation, so a higher concentration of agonist is needed to occupy the required number of receptors. From the concentration–response curves in Figure 2, we can conclude that the concentration of receptor (RT) in a tissue is a key determinant of the relationship between agonist concentration and response. In addition, there is a larger receptor reserve for agonist R than agonist B, and, consequently, agonist R has a higher efficacy than agonist B, even though in the untreated tissue agonists R and B produce the same Emax. Treatment of rats with irreversible inactivators of MOP receptors have shown, for example, that for opioid-induced analgesia, there is a larger receptor reserve for [D-Ala2, NMe-Phe4, Gly-ol5]-enkephalin (DAMGO) than for morphine (Mjanger and Yaksh, 1991), indicating that DAMGO has higher efficacy than morphine for this MOP receptor-mediated response. The inactivation technique can also be used in other ways. Irreversible inactivators have been used to demonstrate that chronic morphine treatment, presumably via some form of desensitization, reduces the receptor reserve for MOP receptor agonist-induced responses (Chavkin and Goldstein, 1984; Christie et al., 1987).

    figure

    Agonists can have different receptor reserves for a response. (A) Diagrammatic representation of two agonists, R and B, which act at the same receptor subtype and are full agonists for the response measured. (B) Following pretreatment of the tissue with a low concentration of an irreversible antagonist (e.g. β-funaltrexamine for MOP receptors) the curve for agonist R shifts to the right but the maximum response stays the same. Pretreatment with a higher concentration of the irreversible antagonist further shifts the curve to the right but also reduces the maximum response by about 35%. (C) Pretreatments of the tissue with the irreversible antagonist produce little shift of agonist B, but the maximum response is markedly reduced, with almost no response recordable after the higher concentration of irreversible antagonist. The receptor reserve for agonist R must be significantly greater than the one for agonist B, with the receptor reserve for agonist B being very small as there is a little rightward shift of the curve for the agonist before the maximum response decreases.

    Further accessible discussions of efficacy, receptor reserve and related issues are available in the literature (Ruffolo, 1982; Clarke and Bond, 1998; Christopoulos and El-Fakahany, 1999; Strange, 2008; Clarke and Berg, 2010; Stallaert et al., 2011).

    Intrinsic activity

    Common misconceptions among those who attempt to analyse drug action are that efficacy can always be directly measured from the maximum response to the drug (although note that this is possible in some cases), and that different drugs producing the same maximum response in a tissue have the same value of efficacy. These misunderstandings are surprisingly common and can lead to incorrect conclusions when investigating agonist action. In fact, the maximum response that an agonist produces in a tissue relative to a full agonist is correctly referred to as the intrinsic activity of the drug. If two drugs are full agonists in a tissue, then they are said to each have an intrinsic activity of 1. If a drug is a partial agonist (e.g. the purple agonist curve in Figure 1B) and produces a maximum response that is 65% of that of a full agonist, it is said to have an intrinsic activity of 0.65. In general, intrinsic activity does give an indication of efficacy when comparing partial agonists, but it is of little use in the analysis of full agonists because all such agonists will have an intrinsic activity of 1, even though they may well have very different values of efficacy.

    Intrinsic efficacy

    Efficacy itself is composed of drug-dependent and tissue-dependent components. The drug-dependent component is referred to as the intrinsic efficacy, which is the ability of the agonist drug molecule, once bound, to activate the receptor protein, or, put another way, the tendency of the drug to increase the proportion of the receptor population in an active conformation. The tissue-dependent components of efficacy include the total receptor concentration and the efficiency of coupling the receptor activation to the measured tissue response. These terms are described in a classic equation based mainly on the work of Stephenson (1956) and Furchgott and Bursztyn (1967):
    urn:x-wiley:00071188:media:bph12222:bph12222-math-0001(1)
    where E is the agonist response, Emax is the maximum tissue response to the agonist, f is some function of the signal produced by the binding of drug to receptor which represents the efficiency of coupling receptor to response, ε is the intrinsic efficacy, RT is the receptor concentration, [A] is the agonist concentration and KA is the equilibrium dissociation constant of the drug–receptor interaction. In the equation, ε and KA are drug-dependent factors, while RT and f are tissue-dependent factors.

    In theory, the relative intrinsic efficacy of one agonist to another for a particular signalling response should be the same irrespective of the tissue where the receptor is expressed. However, even though the intrinsic efficacy may be unchanged, the overall efficacy and hence potency and maximum response (also intrinsic activity) of a drug can vary from tissue to tissue because the factors f and/or RT, as described in Equation 1, can vary from tissue to tissue (Figure 3). For an example of this, see figure 1 of Selley et al. (1997), where MOP receptor agonist responses are compared in rat thalamus and CHO cells expressing MOP receptors.

    figure

    Efficacy is tissue-dependent. Diagrammatic representation of a family of concentration–response curves obtained from different tissues for a response produced by an agonist acting at one particular receptor subtype. Although the agonist has the same value of intrinsic efficacy for the receptor response, variations in the tissue-dependent factors of receptor concentration (RT) and/or efficiency of coupling of receptor to response can markedly affect the potency and maximum response of the agonist. Note also that decreasing τ from tissue 1→5 would also produce the same family of curves.

    In many cases, these considerations do not receive sufficient attention when considering agonist action and often it is assumed, for example, that a full agonist in one tissue will be a full agonist in another, or a partial agonist in one tissue will behave in a similar fashion in all tissues and for all responses examined. In fact, a drug that is a full agonist in one tissue may, due to lower RT or efficiency of coupling receptor activation to response, have lower potency in another tissue, or may be a partial agonist or even an antagonist in another (Figure 3). For example, morphine is a full agonist for inhibition of adenylyl cyclase activity in HEK293 cells (Zaki et al., 2000), yet is an extremely weak partial agonist/competitive antagonist in the rat vas deferens preparation (Ishii et al., 1981; Smith and Rance, 1983). In some ways, the situation is not helped by pharmacological databases such as the IUPHAR database (http://www.iuphar-db.org/index.jsp), where morphine is defined as a full agonist at MOP receptors, but whether or not it behaves as a full or partial agonist depends totally upon the assay/tissue being used. This means that it is very important to describe efficacy in terms of both the tissue and the response measured (see Table 1 as an example).

    Table 1. Intrinsic efficacy of morphine relative to DAMGO (the latter taken as 1 in each case) at MOP receptors for assays involving G protein-mediated responses
    Study Tissue Assay Analysis Efficacy of morphine relative to DAMGO
    Emmerson et al. (1996) C6 glioma cells, rat receptor (heterologous) [35S]GTPγS binding Ehlerta 0.27
    Selley et al. (1998) CHO cells, mouse receptor (heterologous) [35S]GTPγS binding Ehlerta 0.72
    Rat thalamus (endogenous) [35S]GTPγS binding 0.56
    Borgland et al. (2003) AtT20 cells, mouse receptor (heterologous) Ca2+ channel inhibition Nullb,c, b,c 0.58
    Midpoint shiftc 0.74
    Rónai et al. (2006) Mouse vas deferens (endogenous) Inhibition of field-stimulated contraction Nullb 0.28
    Operational 0.07
    McPherson et al. (2010) HEK293 cells, rat receptor (heterologous) [35S]GTPγS binding Operational 0.18
    Nickolls et al. (2011) U2OS osteosarcoma cells, human receptor (heterologous) [35S]GTPγS binding Operational 0.57
    cAMP inhibition 0.27
    Rivero et al. (2012) Rat locus coeruleus neurones (endogenous) K+ current activation Operational 0.02
    Madia et al. (2012) Rat spinal cord (endogenous) [35S]GTPγS binding Operational 0.07
    • a As analysed by Equation 2 above.
    • b Based on the method of Furchgott and Bursztyn (1967).
    • c As described in Borgland et al. (2003).
    • Taking all the values in the table, the intrinsic efficacy of morphine was 0.36 ± 0.07 (mean ± SEM) relative to DAMGO; if only the results from operational analysis are included, this becomes 0.19 ± 0.08, and if only the results from the non-operational analysis are included this becomes 0.52 ± 0.08.

    How is drug efficacy quantified?

    There are a number of approaches used to determine efficacy (Furchgott and Bursztyn, 1967; Black and Leff, 1983; Ehlert, 1985, 2008; Figueroa et al., 2009; Kenakin et al., 2012; Kenakin and Christopoulos, 2013), or more specifically, the relative intrinsic efficacy of a drug in a tissue (i.e. intrinsic efficacy is usually measured relative to other agonists acting at the same receptor in the same tissue rather than as an absolute quantity for each agonist). Furchgott and Bursztyn (1967) developed a receptor inactivation method that compared agonist concentrations that evoked the same level of response before or after inactivation of a fraction of the total available receptors with an irreversible antagonist. From this, the agonist affinity could be calculated, which was then used to construct log occupancy–response curves, with the relative shift between the curves being taken as a measure of relative intrinsic efficacy; the greater the shift, the greater the difference in agonist intrinsic efficacy. Because the measurements were being made at the same population of receptors in a tissue and because differences in affinity have been accounted for by calculating fractional receptor occupancy, then all tissue-dependent factors (i.e. RT and f) cancel out, and therefore, differences between the agonists must be due to differences in the intrinsic efficacy, ε.

    More simply, it has been proposed by Ehlert (1985) and Strange (2008) that the efficacy of one agonist relative to another can be determined from a simple combination of the difference between KA and EC50, and the maximum response. As described by Ehlert (1985):
    urn:x-wiley:00071188:media:bph12222:bph12222-math-0002(2)
    where e is the efficacy of the test agonist, and urn:x-wiley:00071188:media:bph12222:bph12222-math-5001 are the relative maximum response values of the test agonist and an agonist giving a full response, while KA and EC50 are the equilibrium dissociation constant and EC50, respectively, of the test agonist.

    For the analysis in Equation 2, if efficacy measurements are made for a series of agonists producing a response by acting at the same receptor population in a tissue, then the relative efficacy values (e) calculated become relative intrinsic efficacy values (ε). This approach, which has been used to estimate the relative efficacy of agonists at MOP receptors (Emmerson et al., 1996; Selley et al., 1998; Rivero et al., 2012), is more staightforward than Furchgott's as the widespread availablility of radioligands makes the measurement of KA, which is necessary for the calculation in Equation 2, a simple procedure. However, as the KA value obtained from a membrane preparation can vary depending upon the presence or absence of, for example, guanine nucleotide or Na+ ions, then whether a particular KA value obtained under such conditions is the appropriate one to use to calculate efficacy is another matter. This issue is discussed further below.

    An important method developed by Black and Leff (1983) to analyse agonist action is the operational model of agonism, where data from agonist concentration–response curves are fitted to the following equation:
    urn:x-wiley:00071188:media:bph12222:bph12222-math-0003(3)
    where E is the agonist response, Emax is the maximum possible tissue response to the agonist (which can be greater than the measured Emax), n is the slope factor of the transducer function, [A] is the agonist concentration, KA is the equilibrium dissociation constant of the drug–receptor interaction and τ is the operational efficacy, which is RT/Ke, where RT is the receptor concentration and Ke is the concentration of the agonist–receptor complex that produces a half-maximal response (note Ke is not the same thing as the EC50).

    The τ value, or transducer function, is also referred to as the operational efficacy of the agonist and is the ratio of RT/Ke. Although τ has both tissue-dependent and tissue-independent components, if relative τ values are calculated for agonist responses at a receptor population in a cell type or tissue, then the tissue-dependent components cancel out and relative τ becomes a measure of relative intrinsic efficacy. The τ values for agonists can be calculated by simultaneously fitting all agonist concentration–effect curves to the equation, using KA values obtained from separate binding assays. Alternatively, the τ value of an agonist can be obtained by fitting the curve for agonist before and after inactivation of a fraction of receptors. In this case, it is not necessary to first determine KA as the fitting will give a value of this parameter; this may be an advantage if there is concern about using KA values determined from separate binding assays. In summary, the operational model represents an important and increasingly popular approach to quantify agonist action, and indeed Black and Leff's approach has already been used to analyse MOP receptor agonist efficacy and MOP receptor desensitization in a number of experimental settings (Osborne and Williams, 1995; Cox et al., 1998; Garrido et al., 2000; Rónai et al., 2006; Bailey et al., 2009; McPherson et al., 2010; Nickolls et al., 2011; Madia et al., 2012; Rivero et al., 2012).

    Ligand bias

    There is enormous interest in the phenomenon of biased agonism (also known as functional selectivity, protean agonism and ligand-directed signalling; Urban et al., 2007), not least because it offers ways in which more effective drug treatments can be developed. For example, a biased agonist at a receptor might activate a signalling pathway that leads to the desired clinical response but would not activate a signalling pathway responsible for adverse effects (Rajagopal et al., 2010; Whalen et al., 2011), whereas an unbiased agonist would activate both pathways and produce both desired and adverse effects. In many cases, bias is described in terms of signalling via G protein versus arrestin, although bias, in fact, refers to any signalling pathway including those involving different subtypes of G protein or arrestin. Currently, in the field of GPCR research, there is accumulating evidence that bias is widespread (Kahsai et al., 2011; but see Langemeijer et al., 2013). Whatever the case, the measurement of agonist efficacy is of central importance when trying to determine possible ligand bias for signalling pathways. This is, in part, because bias is usually not absolute and it is probably rare for a biased ligand at a GPCR to fully activate one pathway and be completely inactive at another. Accordingly, bias may not be obvious from cursory inspection of concentration–response curves, while differences in receptor reserve for distinct responses emanating from the same receptor can produce data that resemble bias, but closer analysis reveals otherwise. Therefore, rigorous methods to determine efficacy for different signalling pathways are needed, as well as methods to quantify any bias that may be present. Such approaches have now been developed (Rajagopal et al., 2011; Kenakin et al., 2012) and implemented (Evans et al., 2011; Nijmeijer et al., 2012; Rivero et al., 2012; Wacker et al., 2013). It will be of interest to see whether these different approaches lead to similar conclusions with regard to the bias displayed by particular agonists at a GPCR.

    An important recent development is the proposal that analysis of concentration–response curves alone can provide adequate quantification of agonist signalling in order to determine ligand bias. These measures, which include components of agonist affinity and efficacy, include the ‘intrinsic relative activity’ (RAi; not to be confused with intrinsic activity as described earlier) as developed by Ehlert (Griffin et al., 2007; Ehlert, 2008; Figueroa et al., 2009), and the ‘transduction coefficient’ or ‘transduction ratio’, which is the ratio τ/KA as developed by Kenakin and Christopoulos (Kenakin et al., 2012; Kenakin and Christopoulos, 2013). These quantities can be obtained by direct fitting of agonist concentration–response curves to the relevant equations. A significant advantage of these approaches is that independent measures of ligand affinity or partial inactivation of the receptor population is not required to obtain the measures of agonist signalling (Griffin et al., 2007; Ehlert, 2008; Figueroa et al., 2009; Stallaert et al., 2011; Kenakin et al., 2012; Kenakin and Christopoulos, 2013).

    These approaches are constantly evolving (Ehlert et al., 2011; Kenakin, 2013; Kenakin and Christopoulos, 2013), and although the theoretical basis can seem complex, such ideas are likely to see increased application in the future because of their importance in assessing ligand bias. Whichever approach is used to determine ligand bias at a receptor, with modern computer packages now available, and with practice and advice (e.g. Hall and Langmead, 2010), it should be possible for the experimenter to fit data to these models.

    Context-dependent efficacy

    Another important consideration is that of context-dependent efficacy. This relates in many ways to the efficiency of coupling. Here, the efficacy is dependent upon the presence of other factors that may be present or absent in a cell, or present at very different concentrations, such as particular proteins that interact with the receptor. A clear example of this is receptor phosphorylation. Thus, the efficacy of arrestin recruitment is, in most cases, increased by receptor phosphorylation by G protein coupled receptor kinases (GRKs). Accordingly, when arrestin-dependent signalling is being examined, the ligand efficacy will be greatly influenced by the concentration of the relevant GRK and phosphatase subtypes present in the cell type, or even subcellular compartment of the cell. A good example of this is morphine's ability to induce MOP receptor trafficking. In the vast majority of cell types, morphine produces little or no MOP receptor internalization, unless GRK2 is overexpressed (Zhang et al., 1998; Schulz et al., 2004). However, in certain cell types, such as striatal neurones (Haberstock-Debic et al., 2005), or in certain neuronal compartments, such as the dendrites of nucleus accumbens neurones (Haberstock-Debic et al., 2003), morphine is able to induce significant trafficking without the need for overexpression of GRKs. Other ways in which cellular context could affect overall efficacy and possibly even the relative intrinsic efficacy of agonists are discussed below (see also Figure 9).

    Agonist efficacy for G protein signalling at MOP receptors

    The complexities of opioid action in the intact organism mean that it is very important to establish opioid drug efficacy at the receptor (Kelly et al., 2008; Morgan and Christie, 2011), not least because the relative intrinsic efficacy of MOP receptor agonists obtained for complex responses, such as analgesia, is a product of pharmacokinetics, the selectivity of ligands for opioid receptor subtypes, as well as agonist action at MOP receptors and the immediate post-receptor environment. Relative intrinsic efficacy values are also extremely useful when studying MOP receptor desensitization and its relationship to tolerance as changes in intrinsic efficacy as seen in desensitization are a more convenient way to quantify desensitization (Navratilova et al., 2007; Bailey et al., 2009) than having to deal with often complex changes in agonist maximum response and potency. Last but not least, reliable measures of intrinsic efficacy are needed to detect and quantify ligand bias, as described below.

    Relative intrinsic activity

    In the majority of cases, estimations of the efficacy of MOP receptor ligands have been obtained from measures of relative intrinsic activity (Whistler et al., 1999; Molinari et al., 2010). In some cases, this provides a reasonable estimation of efficacy, for example, in [35S]GTPγS binding assays using brain membranes, where the receptor reserve tends to be small and a good number of ligands behave as partial agonists (Selley et al., 1998). On the other hand, when using [35S]GTPγS assays with clonal cell lines stably expressing reasonably high levels of receptor, many ligands with lower efficacy behave as full agonists, and thus, intrinsic activity is not a good measure of intrinsic efficacy. Also, for assays where there is a high level of receptor reserve associated with the response, such as MOP receptor-induced inhibition of AC activity, intrinsic activity as a means to estimate ligand efficacy is unable to distinguish high efficacy agonists from agonists that are weak partial agonists in other assay systems. Nevertheless, the use of intrinsic activity to compare MOP receptor agonists remains in widespread use (Molinari et al., 2010; Frölich et al., 2011).

    Relative intrinsic efficacy

    To determine relative intrinsic efficacy at MOP receptors, we assessed agonist-induced stimulation of [35S]GTPγS binding to membranes of HEK293 cells stably expressing MOP receptors at a moderately high density of 778 fmol·mg−1 protein (McPherson et al., 2010). Full concentration–effect curves were constructed for 22 ligands. In addition, to use operational analysis without receptor inactivation, the binding constant (equilibrium dissociation constants, KA) for each agonist was determined. Using these data, the concentration–response curves were simultaneously fitted in GraphPad Prism (GraphPad Software Inc., La Jolla, CA, USA) to the operational model, as in Equation 3 above. By sharing the values of Emax and n for all agonists, and inserting KA values for each agonist (this was achieved by using the respective KA value as the column heading for each agonist in the GraphPad Prism data table), fitted values of τ were obtained. DAMGO had the highest τ in this assay (28.5), with Met-enkephalin and methadone also having high values (22.0 and 18.2, respectively); morphine had a much lower τ value (5.2), while opioids such as buprenorphine and pentazocine had extremely low τ values (0.6 and 0.3 respectively). To determine whether the τ values of operational efficacy we obtained in HEK293 cells might reflect those for a G protein-mediated response for endogenous MOP receptors expressed in neurones, we also determined τ values for agonists at MOP receptors expressed in locus coeruleus neurones contained in slices of rat midbrain. The response involved MOP receptor-induced activation of an inwardly rectifying K+ current recorded in whole cell patch-clamp mode (Rivero et al., 2012). For this analysis, we constructed concentration–response curves to agonists in the absence or presence of the irreversible MOP receptor inactivator β-FNA. The pairs of curves for each agonist were again analysed by operational modelling, but this time with KA, Em and n values constrained to be shared for each agonist curve before and after receptor inactivation. The relative τ values obtained correlated well with those from HEK293 cells (Figure 4), indicating that relative (not absolute) τ values for agonists at a receptor appear to be tissue-independent, and that in this case at least the relative τ values obtained in a clonal cell line represent those that might be obtained for endogenous receptors in neurones.

    figure

    Relative intrinsic efficacy (τ) values for agonists at MOP receptors are tissue-independent. Correlation of τ values for [35S]GTPγS binding in HEK293 cells and activation of a K+ current in locus coeruleus neurones for four MOP receptor agonists. Following linear regression, the r2 value was 0.975, indicating a high correlation of agonist relative τ values obtained in the two systems. The analysis is based on data from McPherson et al. (2010) and Rivero et al. (2012).

    It is interesting to compare the relative intrinsic efficacy values obtained in these experiments to those obtained in other studies of MOP receptor-mediated G protein-dependent responses (Table 1). The agonists DAMGO and morphine have been used in all the cited studies, and it can be seen that the intrinsic efficacy of morphine relative to DAMGO is always significantly less than 1, although there is some variation depending upon the method used. In general, operational analysis results in lower values of intrinsic efficacy for morphine relative to DAMGO, but the reason why morphine's intrinsic efficacy relative to DAMGO might be either underestimated in operational analysis and/or overestimated in other types of analysis is unclear.

    As mentioned previously, an important further issue relates to the KA values used in the operational analysis. In our study, we determined KA values in the presence of a high concentration of Na+ (137 mM), which is known to stabilize what is likely to be a basal state of the receptor (Strange, 2008). Whether these KA values are the appropriate ones to use in the operational analysis is an assumption that remains open to question. We used the KA values determined as above on the basis that since the KA values of all 22 ligands in our study were obtained by the same method, then on balance this would not significantly affect the overall relative operational efficacy measurements. However, in a recent review, Kenakin and Christopoulos (2013) present evidence that it may not be possible to use a single KA value obtained from a binding assay to fit data for a single agonist from different functional assays to the operational model. Furthermore, when we used the operational model itself to determine KA values from concentration–response curves before and after receptor inactivation (Rivero et al., 2012), we found the KA values obtained to be quite different from those obtained in binding assays. As the KA values obtained from the operational fitting can be regarded as macroscopic binding constants and those from ligand binding assays as microscopic constants, then it is perhaps not surprising that they are different and therefore should not be directly compared. These concerns about KA and its use in this type of analysis remain unresolved (Colquhoun, 1998) and are the subject of recent and continuing discussion (Kenakin, 2013; Kenakin and Christopoulos, 2013).

    Agonist efficacy for other signalling outputs from MOP receptors

    Apart from G protein-mediated responses, we also wanted to determine the relative intrinsic efficacy of MOP receptor ligands for other signalling events, including arrestin-mediated responses. Apart from mediating the uncoupling of GPCR and G protein, the non-visual arrestins (arrestin-2 and -3) are known to interact with and regulate the function of a large number of signalling proteins such as elements of the MAPK pathway (Luttrell and Gesty-Palmer, 2010; see also Figure 5A). Until relatively recently, quantifying arrestin function has been problematic because of the absence of convenient assays to determine agonist concentration–response relationships for arrestin interaction with a GPCR. However, commercially available arrestin recruitment assays are now available (Bassoni et al., 2012). In addition, the development of FRET (Frölich et al., 2011), BRET (Molinari et al., 2010) and bimolecular fluorescence complementation (Kilpatrick and Holliday, 2012) techniques has enabled the rapid collection of GPCR-arrestin recruitment data. Our own approach (McPherson et al., 2010) was to use the Pathunter system from DiscoveRx (DiscoveRx Corporation, Birmingham, UK), which is a complementation assay where different parts of the enzyme β-galactosidase are fused to either the C-terminal tail of the MOP receptor or to arrestin-3. With this assay, we produced concentration–response curves for all the agonists used in the [35S]GTPγS binding assay described above, and the curves were again subjected to operational analysis and τ values calculated. The absolute τ values were much lower than those for [35S]GTPγS binding but this is not surprising since the potency and/or maximum responses were in all cases lower for the arrestin assay than for G protein coupling. This probably reflects the lack of amplification and hence receptor reserve in the arrestin assay compared to the [35S]GTPγS binding assay. In addition, we were able to use operational analysis to determine the relative intrinsic efficacy for MOP receptor phosphorylation on Ser375 in the C-terminal tail of the receptor. Phosphorylation of this residue, presumably by a G protein coupled receptor kinase, is a key step in MOP receptor desensitization and trafficking (Schulz et al., 2004; Doll et al., 2012), and a commercially available anti-phosphoreceptor antibody for phosphorylated Ser375 has been developed. We constructed full agonist concentration–phosphorylation curves for four agonists, and the pattern of relative τ values we obtained for Ser375 phosphorylation (DAMGO ≥ etorphine ≥ endomorphin-2 > morphine) was the same as that for arrestin-3 recruitment. Absolute τ values were again low and similar to those obtained for arrestin recruitment. Thus, we were able to use operational analysis to determine relative intrinsic efficacy for multiple signalling outputs from a single receptor subtype. The further manipulation and interpretation of these data in relation to biased agonism is discussed below.

    figure

    (A) Major cellular signalling pathways activated by MOP receptor (MOPr) agonists. While the G protein-dependent signalling pathways have been established over many years, the arrestin-dependent signalling pathways have only recently begun to be characterized (Walwyn et al., 2007; Groer et al., 2011; Raehal et al., 2011; Henry et al., 2012). The MAPK cascade can be activated via G protein- or arrestin-dependent pathways (Zheng et al., 2008). Many of these pathways also lead to changes in gene expression, particularly with prolonged agonist treatments. (B) The in vivo consequences of MOP receptor agonist administration to a patient. Effects in red are desirable, therapeutic effects; those in blue are adverse, undesirable effects. Some, such as sedation or euphoria, can be therapeutic or undesirable depending upon the context.

    User-friendliness and suitability of the operational model

    How easy is it to fit data to the operational model and obtain reliable values of τ? Operational modelling requires good concentration–response data, particularly if the analysis involves only a small number of ligands; tight data and well-defined maxima all help. By using constraints, such as of the n value or Emax value, it can be possible to fit data that at first attempt appears difficult to fit. Often it is helpful to first fit the data to a logistic model (sigmoidal concentration–response) to obtain estimates of maximum response and n that can be used as initial estimates in the subsequent operational analysis (note, however, that Em and n do not have exactly the same meaning in the logistic and operational models, but the values obtained from fitting to the logistic model are fine to use as starting points for operational fitting). Both logistic and operational fitting can be undertaken in a suitable data analysis programme such as GraphPad Prism, which already includes the equation for the operational model. However, even if the fitting of the data is able to produce τ values, care must be taken to check that the errors associated with fitting are not so large as to make the τ values essentially meaningless. In this regard, our attempts to analyse MOP receptor internalization data using operational modelling were unsuccessful and so we used the approach described by Ehlert, (1985) to estimate relative intrinsic efficacy for this readout (Equation 2 above).

    An important question is whether the operational model is actually suitable to analyse concentration–response data for responses (such as arrestin recruitment and receptor phosphorylation) that have little or no receptor reserve. It has previously been suggested that data showing a linear relationship between occupancy and response should not be fitted to the operational model (Nickolls et al., 2011) as the latter relies upon a model where the occupancy–response relationship is hyperbolic. To determine the relationship between occupancy and response, a straightforward approach is to calculate fractional receptor occupancy at each concentration of ligand using the simple relationship:
    urn:x-wiley:00071188:media:bph12222:bph12222-math-0004(4)
    where ρ is the fractional receptor occupancy, [A] is the concentration of agonist and KA is the equilibrium dissociation constant. Fitting of the resulting data, for example, to a linear regression or hyperbolic function reveals the nature of the occupancy–response relationship.

    For the MOP receptor data described above from Rivero et al. (2012), we found that the relationships between agonist fractional receptor occupancy and [35S]GTPγS binding are best described by hyperbolic curves (Figure 6A; see also Figure 7 and Supplementary Figures S3 and S4 from Rivero et al. 2012), indicating amplification and a receptor reserve. On the other hand, the relationships between receptor occupancy and either arrestin recruitment, Ser375 phosphorylation or internalization did not fit well or could not be fitted to a hyperbolic relationship. Rather, the data fitted better to a linear relationship than a hyperbolic one (Figure 6). This is perhaps not surprising because there is unlikely to be much amplification in these signalling readouts, and these responses normally require high levels of receptor occupancy (Borgland et al., 2003). However, it should be noted that the relationship between receptor occupancy and the regulatory readouts may be more complex yet, as, for example, these data could also be readily fitted to an exponential growth model (Figure 6E,F). It is not immediately obvious what this might mean, but possibilities include that there is a high occupancy threshold for these regulatory responses, or that hierarchical phosphorylation (Lau et al., 2011; Just et al., 2013) of the C-terminus of MOP receptors results in a complex occupancy–response relationship, which is accordingly reflected in the processes of arrestin recruitment and receptor internalization.

    figure

    Occupancy–response relationships for DAMGO and morphine for (A) [35S]GTPγS binding fitted to a hyperbolic function, for (B–D) MOP receptor internalization, Ser375 phosphorylation and arrestin-3 recruitment fitted by linear regression, and for (E-F) the Ser375 and arrestin-3 data points from C and D instead fitted to an exponential growth model. The values of occupancy were calculated using Equation 4, using data previously published (McPherson et al., 2010; Rivero et al., 2012). The values shown in brackets are the R2 values from the fits in GraphPad Prism. For linear regression, in all cases the line was constrained to go through X = 0, Y = 0. The occupancy–response relationships for Ser375 phosphorylation, arrestin-3 recruitment and internalization were better described by linear than hyperbolic relationships, indicating little or no receptor reserve for these responses in this assay system (R2 values for fitting of occupany–response data for Ser375 to a hyperbolic function was 0.750 for DAMGO and did not converge for morphine; for arrestin-3 recruitment 0.788 for DAMGO and 0.704 for morphine; for internalization 0.874 for DAMGO and 0.658 for morphine). However, the R2 values for fitting Ser375 phosphorylation (E), arrestin-3 recruitment (F) and internalization (not shown) to an exponential growth model [Y = Y0 × exp(k × X)] was, in turn, better than to a linear model.

    figure

    Correlation of efficacy values for [35S]GTPγS and arrestin recruitment for a series of MOP receptor agonists. In each case, the data are expressed as a fraction of the value for DAMGO (taken as 1 in each case), and were subjected to linear regression. (A) Correlation of τ values with linear regression. (B) Correlation using intrinsic activity values for arrestin recruitment and τ values for [35S]GTPγS binding, with linear regression. Note that the relationship when using intrinsic activity values for arrestin recruitment is essentially the same as when using τ values for this pathway as in A. (C) Correlation using intrinsic activity values for [35S]GTPγS binding and τ values for arrestin recruitment, with linear regression. The use of intrinsic activity for G protein output now radically alters the relationship compared to graphs A and B. Note also that endomorphin-2 no longer appears as an arrestin-biased agonist, and in fact now it almost overlies DAMGO in the correlation.

    It may well be that for the three regulatory readouts of phosphorylation, arrestin recruitment and internalization, the Emax value of each agonist (i.e. the intrinsic activity) is sufficient as a measure of intrinsic efficacy, as discussed in the section below (also Figure 7A,B). In summary, for responses with no receptor reserve, although operational analysis evidently does provide reasonable estimates of relative intrinsic efficacy (McPherson et al., 2010), it may be an overcomplicated way to treat the data, especially as relative intrinsic activity (as given by relative Emax values) is easier to measure.

    Biased agonism at MOP receptors

    An important reason to obtain accurate measures of agonist relative intrinsic efficacy is to facilitate detection of ligand bias. A large number of ligands exist for MOP receptors and many of the responses to MOP receptor activation, such as ion channel modulation and adenylyl cyclase inhibition are known to be mediated by G proteins of the Gi/o subfamily (Figure 5A). On the other hand, the role of arrestin-dependent signalling through MOP receptors is largely unknown, although responses to MOP receptor agonists are significantly modified in arrestin-3 knockout mice (Bohn et al., 1999; Raehal et al., 2005, 2011). Given the importance of this receptor in pain and reward pathways, it is important to evaluate possible ligand bias as a means to explore the role of different signalling pathways in opioid drug action in vivo (Figure 5B). For example, could a G protein-biased or arrestin-biased opioid agonist be an effective analgesic but display fewer adverse effects such as dependence? To determine bias, we initially constructed a correlation plot of τ values for arrestin-3 recruitment versus τ values of [35S]GTPγS binding (Figure 7A). When the experimental data were subjected to linear regression, the overall correlation was quite high (r2 = 0.646 for the entire series of MOP receptor ligands; McPherson et al., 2010), indicating that in general, the better the agonist is in promoting coupling to G protein, the better the agonist is at recruiting arrestin. This was not unexpected and has been reported for other GPCRs such as the β2-adrenoceptor (January et al., 1997). Interestingly, and of relevance to the preceding section, if intrinsic activity values are used instead of τ values for arrestin recruitment, the same relationship is obtained (Figure 7B). However, the same does not hold if intrinsic activity values are used instead of τ values for [35S]GTPγS binding (Figure 7C) because most of the agonists examined were full agonists in this assay and hence had an intrinsic activity of close to 1. Further inspection of the correlation in Figure 7A indicated two points of particular interest. Firstly, it appears that the endomorphins are biased towards arrestin recruitment, and, secondly, that morphine is not biased between G protein and arrestin pathways.

    The endomorphins are arrestin-biased

    That the endomorphins are arrestin-biased was an unexpected finding as the tetrapeptides endomorphin-1 and -2 (Zadina et al., 1997) had not previously been identified as biased ligands. Further quantitative analysis of the data by a recently published method (Rajagopal et al., 2011) indicated that endomorphin-2 was significantly biased towards arrestin recruitment over G protein activation, although it should be noted that a number of the agonists, such as etorphine and alfentanil, appeared to show at least some arrestin bias (see figure 8 in Rivero et al. 2012). The approach to quantify bias involves calculating the ‘effective signalling’ for a ligand (σlig) by the expression σlig = log(τligref), where the τ value for a particular ligand for a signalling pathway (pathway 1) is compared to that of a reference, unbiased (‘balanced’) ligand in the same signalling pathway. This is repeated for another signalling output (pathway 2), allowing a bias factor (β) to be calculated (Rajagopal et al., 2011):
    urn:x-wiley:00071188:media:bph12222:bph12222-math-0005(5)
    The arrestin bias factor for endomorphin-2 was found to be −0.81 ± 0.18. A recent study where agonist-induced BRET between MOP receptors and G proteins, and MOP receptors and arrestin was investigated, did not identify endomorphin-2 as a biased agonist (Molinari et al., 2010), but this may be because intrinsic activity, rather than intrinsic efficacy, was used in the correlation (this can be clearly seen from our data in Figure 7C). Careful inspection of the Molinari et al. (2010) data indicates that the EC50 values for G protein activation for most MOP receptor agonists studied were significantly lower (∼5–10-fold) than those for arrestin interaction (in line with a low or non-existent receptor reserve for the latter) – apart notably from endomorphin-2, where the potency for the two responses was the same. This strongly suggests that, relative to the other ligands tested, endomorphin-2 is biased towards arrestin recruitment over G protein activation. Further support for arrestin bias for endomorphin-2 comes from older studies, where the endomorphins were able to promote efficient internalization of MOP receptors (McConalogue et al., 1999) yet were reported to be partial agonists in [35S]GTPγS studies (Sim et al., 1998). The finding that endomorphin-2 is an arrestin-biased ligand is of interest as it provides a lead to study the role of arrestin-dependent pathways in opioid receptor action (Figure 5), and also because there has been recent growing interest in ligands based on the structure of the endomorphins as efficient analgesics but with reduced tendency to induce dependence or respiratory depression (Wilson et al., 2000; Varamini et al., 2012).

    Morphine an unbiased ligand?

    Morphine was not identified as a biased ligand in our analysis, whereas it has been previously suggested to be a G protein-biased ligand (Borgland et al., 2003; Raehal et al., 2011). Indeed, early reports emphasized the inability of morphine to induce MOP receptor internalization while at the same time being able to activate G protein-dependent signalling pathways including K+ current activation and cyclic AMP inhibition (Whistler et al., 1999; Zaki et al., 2000). Can it therefore be that morphine is truly unbiased? An important point to grasp here is that just as two agonists producing the same response in a tissue can have different receptor reserves (Figure 2), so an agonist acting at one receptor subtype can produce two responses that have different receptor reserves, which has nothing to do with bias per se. For the MOP receptor agonists in general, the potency with which they activate G protein-coupled responses is greater (i.e. lower EC50) than for regulatory responses (arrestin recruitment, phosphorylation, internalization and desensitization; Borgland et al., 2003; McPherson et al., 2010; Molinari et al., 2010; Rivero et al., 2012). Thus, the agonist receptor reserve (a reflection of efficacy) for G protein coupling is greater than that for the regulatory responses, meaning that for most agonists, the concentration response curve for G protein coupling lies to the left of that for, say, arrestin recruitment (Figure 8). For a high efficacy agonist such as DAMGO, it is possible to achieve maximum tissue responses for both assays. However, for a lower efficacy agonist like morphine, although it has sufficient efficacy to achieve a maximum response in the [35S]GTPγS binding assay in the particular assay system that we employed, it does not have sufficient efficacy to achieve anything like the maximum response in the arrestin recruitment assay. Visual inspection of the concentration–response curves, however, could lead one to conclude that morphine is biased towards the G protein response, as it appears such a weak agonist at the arrestin recruitment response relative to the G protein response (ironically our quantification of bias actually suggests that DAMGO is somewhat G protein biased; see Figure 7 in this review, also figure 8 in Rivero et al., 2012). However, the important thing to compare is the ratio of intrinsic efficacy values (τ or whatever else is used) for the two different responses measured for each agonist. Using the τ values calculated previously (McPherson et al., 2010), it can be seen that for many of the agonists investigated in that study, the τG protein : τarrestin ratio is >20 for most agonists, including morphine. On this basis, we would not conclude that morphine is different from most of the other agonists tested, that is, morphine's actions in these assay systems can be explained by its relatively low efficacy for the two receptor responses measured. Further, we would predict that for an unbiased agonist with even lower efficacy for these responses than morphine, the agonist would be a partial one for [35S]GTPγS binding but lack detectable agonist activity for arrestin-3 recruitment. This is exactly what is seen, for example, with buprenorphine, pentazocine and meptazinol (McPherson et al., 2010).

    figure

    Comparison of the effects of DAMGO and morphine on ligand binding, [35S]GTPγS binding and arrestin-3 recruitment. (A) DAMGO and morphine displaced [3H]naloxone from MOP receptors with similar potency. (B) Concentration–response curves for activation of [35S]GTPγS binding and arrestin-3 recruitment for MOP receptors stably expressed in HEK293 cells. DAMGO was a full agonist in both assays, whereas morphine was a full agonist in the G protein assay but a weak partial agonist in the arrestin assay. The results in panel (A) indicate that differences between the DAMGO and morphine curves shown in (B) cannot be due to differences in affinity for MOP receptors.

    figure

    Possible mechanism to explain changes in relative intrinsic efficacy of MOP receptor agonists in different tissues/cell types. (A) In tissue 1, the high-efficacy agonist DAMGO is better able to promote coupling of MOP receptors via G protein to the downstream signal than the lower efficacy agonist morphine. (B) In tissue 2, which contains significant levels of an ‘interacting protein’, DAMGO but not morphine is able to promote recruitment of the ‘interacting protein’ to the signalling complex. Recruitment of the ‘Interacting protein’ dampens DAMGO-induced signalling, and consequently reduces the efficacy of DAMGO relative to morphine. DAMGO might have this effect either because it induces a conformation of the receptor-signalling complex more favourable to recruitment of the ‘Interacting protein’, or because the efficacy of morphine is not sufficient to recruit the ‘Interacting protein’. In this example, an RGS protein (Psigfogeorgou et al., 2011; Traynor, 2012) could fulfil the role of an ‘Interacting protein’. Furthermore, such dispartities are more likely to occur if responses with different proximities to the receptor are compared. Thus, if G protein activation using [35S]GTPγS was used in one assay, but the more distal readout of, say, an ion channel was used in another, then the latter could be subject to regulation by interacting proteins at more levels than the former response.

    Contrary to the findings described earlier, using MOP receptors expressed in AtT20 cells, Borgland et al. (2003) described morphine as an agonist with relatively high intrinsic efficacy for Ca2+ channel inhibition, which is a G protein-mediated response, yet very low efficacy for internalization in these cells. On this basis, morphine could be classified as a G protein-biased ligand, but this seems inconsistent with data from other systems, such as inhibition of the nerve-evoked contractions in rat vas deferens (considered to be a G protein-mediated response), where morphine is a very weak partial agonist/competitive antagonist compared to other MOP receptor agonists such as DAMGO (Ishii et al., 1981; Smith and Rance, 1983). Furthermore, if morphine is a G protein-biased agonist, it might be expected to display effects similar to those observed with recently identified G protein-biased MOP receptor ligands (discussed below), such as reduced liability to induce tolerance and constipation (Lamb et al., 2012; DeWire et al., 2013), but this is certainly not the case.

    The results from AtT20 cells (Borgland et al., 2003) cannot be easily reconciled with our own data from HEK293 cells, unless one considers the possibility that the relative efficacies of ligands are subject to tissue-specific variation. For example, if morphine and DAMGO recruited distinct regulatory proteins to the G protein/arrestin signalling complex or even downstream of this, then if these protein partners were expressed at varying levels in tissues, this could consequently affect the relative efficacy values. This possibility is depicted in Figure 9, where the ability of morphine and DAMGO to signal is compared in two tissues or cell types. In the first tissue, DAMGO, due to its high intrinsic efficacy, is better able to promote coupling of MOP receptors to the downstream signal than morphine. In the second tissue, DAMGO also recruits an ‘interacting protein’, for example, a regulator of G protein signalling (RGS) protein (Psigfogeorgou et al., 2011; Traynor, 2012), to the signalling complex (this protein may be absent or expressed at a much lower level in the first tissue), which may selectively dampen DAMGO-induced signalling, and consequently reduce the efficacy of DAMGO relative to morphine. DAMGO might have this effect either because it induces a conformation of the receptor-signalling complex more favourable to recruitment of the interacting protein, or because the efficacy of morphine is not sufficient to recruit the interacting protein.

    In summary, the question of whether or not morphine is a biased agonist at MOP receptors remains a controversial one, but given the clinical and experimental importance of this drug, it merits further study in different systems.

    Efficacy, bias and the clinical use of opioids

    What do pharmacological measures of efficacy and bias obtained in cells mean in terms of the clinical use of opioid drugs? This is a far from simple matter because the relative intrinsic efficacy of MOP receptor agonists obtained for complex responses such as analgesia will be a product not only of agonist action at MOP receptors and the immediate post-receptor environment but also pharmacokinetics and the selectivity of ligands for opioid receptor subtypes (Trescot et al., 2008). In addition, the type of pain being treated and the previous history of opioid medication in the patient will all feed into the overall analgesic effectiveness achieved (McQuay, 1999). It is of interest to note, however, that at the level of MOP receptor coupling, there is a large degree of variation in the efficacy for G protein activation, and indeed for other signalling outputs (Figure 7; also McPherson et al., 2010), yet clinicians using these drugs to induce analgesia would not recognize such differences in analgesic efficacy. Indeed, according to the British Pain Society's pamphlet ‘Opioids for persistent pain: Good practice’ (2010), buprenorphine, pentazocine and meperidine are each classified as ‘strong opioids’, whereas they exhibit relatively low efficacy in cell signalling assays. Accordingly, caution is required when extrapolating pharmacological efficacy from a cell-based test system to the possible in vivo effects of an opioid in a patient. Indeed, in spite of all the experimental research on opioids – often 3 or 4 papers per day on PubMed – morphine has been and remains the first choice for the treatment of severe pain such as in cancer (McQuay, 1999; Bennett et al., 2012). However, as noted previously (Hanks et al., 2001), morphine remains the first choice ‘for reasons of familiarity, availability and cost rather than proven superiority’. Indeed, for all the work so far carried out to determine differences in MOP receptor agonist action, the adverse effects produced by the available ligands are much the same, with constipation and nausea being among the most troublesome (see Figure 5B). Even though buprenorphine is sometimes considered a safer opioid due to its ‘ceiling effect’ for inducing respiratory depression (Dahan et al., 2005), deaths due to buprenorphine-induced respiratory depression are still recorded (Mégarbane et al., 2006; Kim et al., 2012). Given the clinical importance of MOP receptor agonists, the sentiments expressed above should nevertheless not deter the experimenter from trying to understand the molecular actions of both established and recently synthesized MOP receptor ligands in terms of efficacy and bias, in the hope of obtaining better opioid drugs for the treatment of pain (Groer et al., 2007; Lamb et al., 2012).

    Concluding remarks

    What are the intracellular signalling pathways that mediate the therapeutic and/or adverse effects of opioid agonists (Figure 5)? Can biased ligands at the MOP receptor be developed that selectively activate the former over the latter? What happens to these signalling pathways following prolonged MOP receptor activation? These remain central questions in the field of opioid receptor pharmacology. It might reasonably be concluded, for example, that G protein-biased MOP receptor ligands are desirable as these will provide effective analgesia with limited tolerance. Certainly, studies with the novel compound herkinorin, which has marked G protein bias and produces little if any arrestin recruitment (Groer et al., 2007; Lamb et al., 2012), suggest that such compounds have therapeutic potential. In addition, the recent identification of a G protein-biased MOP receptor ligand, TRV130 (DeWire et al., 2013), which is effectively analgesic but produces less respiratory depression and constipation than morphine, is of great interest. On the other hand, ligands based on the structure of the endomorphins indicate that such ligands, which appear to be arrestin-biased, also have a desirable profile as they are analgesic yet produce little abuse potential or respiratory depression (Wilson et al., 2000; Varamini et al., 2012). What these studies really indicate is the need to fully understand the respective role of G proteins and arrestins in the action of agonists at MOP receptors, in producing both the therapeutic and the adverse effects of these drugs (Figure 5). However, this is not a trivial task because, for example, data with MOP receptor agonists from arrestin knockout mice have not provided simple conclusions (Raehal et al., 2005; Dang et al., 2011; Quillinan et al., 2011; Kang et al., 2012) perhaps because arrestin deletion results not only in the loss of that protein but also in other adaptive responses in the animal such as upregulation of JNK (Mittal et al., 2012). Whatever the case, a key element in the development of new, potentially biased MOP receptor ligands will be the application of rigorous methods to determine pathway-dependent agonist efficacy and signalling so that reliable estimates of ligand bias can be obtained. This must be a priority in future studies.

    Acknowledgements

    I would like to thank Professor Graeme Henderson for critically reading this manuscript and for his helpful suggestions and comments. The data cited in this review come from studies supported by the MRC UK [Grant G0600943] and the BBSRC [Grant BB/D012902/1].

      Conflict of interest

      The author declares no conflict of interest.